SAFE public projects git trees. - safe/jmp/linux-2.6/blob - tools/perf/builtin-report.c

   1 /*
   2  * builtin-report.c
   3  *
   4  * Builtin report command: Analyze the perf.data input file,
   5  * look up and read DSOs and symbol information and display
   6  * a histogram of results, along various sorting keys.
   7  */
   8 #include "builtin.h"
   9
  10 #include "util/util.h"
  11
  12 #include "util/color.h"
  13 #include "util/list.h"
  14 #include "util/cache.h"
  15 #include "util/rbtree.h"
  16 #include "util/symbol.h"
  17 #include "util/string.h"
  18
  19 #include "perf.h"
  20
  21 #include "util/parse-options.h"
  22 #include "util/parse-events.h"
  23
  24 #define SHOW_KERNEL     1
  25 #define SHOW_USER       2
  26 #define SHOW_HV         4
  27
  28 static char             const *input_name = "perf.data";
  29 static char             *vmlinux = NULL;
  30
  31 static char             default_sort_order[] = "comm,dso";
  32 static char             *sort_order = default_sort_order;
  33
  34 static int              input;
  35 static int              show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
  36
  37 static int              dump_trace = 0;
  38 #define dprintf(x...)   do { if (dump_trace) printf(x); } while (0)
  39
  40 static int              verbose;
  41 static int              full_paths;
  42
  43 static unsigned long    page_size;
  44 static unsigned long    mmap_window = 32;
  45
  46 struct ip_event {
  47         struct perf_event_header header;
  48         __u64 ip;
  49         __u32 pid, tid;
  50 };
  51
  52 struct mmap_event {
  53         struct perf_event_header header;
  54         __u32 pid, tid;
  55         __u64 start;
  56         __u64 len;
  57         __u64 pgoff;
  58         char filename[PATH_MAX];
  59 };
  60
  61 struct comm_event {
  62         struct perf_event_header header;
  63         __u32 pid, tid;
  64         char comm[16];
  65 };
  66
  67 struct fork_event {
  68         struct perf_event_header header;
  69         __u32 pid, ppid;
  70 };
  71
  72 struct period_event {
  73         struct perf_event_header header;
  74         __u64 time;
  75         __u64 id;
  76         __u64 sample_period;
  77 };
  78
  79 typedef union event_union {
  80         struct perf_event_header        header;
  81         struct ip_event                 ip;
  82         struct mmap_event               mmap;
  83         struct comm_event               comm;
  84         struct fork_event               fork;
  85         struct period_event             period;
  86 } event_t;
  87
  88 static LIST_HEAD(dsos);
  89 static struct dso *kernel_dso;
  90 static struct dso *vdso;
  91
  92 static void dsos__add(struct dso *dso)
  93 {
  94         list_add_tail(&dso->node, &dsos);
  95 }
  96
  97 static struct dso *dsos__find(const char *name)
  98 {
  99         struct dso *pos;
 100
 101         list_for_each_entry(pos, &dsos, node)
 102                 if (strcmp(pos->name, name) == 0)
 103                         return pos;
 104         return NULL;
 105 }
 106
 107 static struct dso *dsos__findnew(const char *name)
 108 {
 109         struct dso *dso = dsos__find(name);
 110         int nr;
 111
 112         if (dso)
 113                 return dso;
 114
 115         dso = dso__new(name, 0);
 116         if (!dso)
 117                 goto out_delete_dso;
 118
 119         nr = dso__load(dso, NULL, verbose);
 120         if (nr < 0) {
 121                 if (verbose)
 122                         fprintf(stderr, "Failed to open: %s\n", name);
 123                 goto out_delete_dso;
 124         }
 125         if (!nr && verbose) {
 126                 fprintf(stderr,
 127                 "No symbols found in: %s, maybe install a debug package?\n",
 128                                 name);
 129         }
 130
 131         dsos__add(dso);
 132
 133         return dso;
 134
 135 out_delete_dso:
 136         dso__delete(dso);
 137         return NULL;
 138 }
 139
 140 static void dsos__fprintf(FILE *fp)
 141 {
 142         struct dso *pos;
 143
 144         list_for_each_entry(pos, &dsos, node)
 145                 dso__fprintf(pos, fp);
 146 }
 147
 148 static struct symbol *vdso__find_symbol(struct dso *dso, uint64_t ip)
 149 {
 150         return dso__find_symbol(kernel_dso, ip);
 151 }
 152
 153 static int load_kernel(void)
 154 {
 155         int err;
 156
 157         kernel_dso = dso__new("[kernel]", 0);
 158         if (!kernel_dso)
 159                 return -1;
 160
 161         err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
 162         if (err) {
 163                 dso__delete(kernel_dso);
 164                 kernel_dso = NULL;
 165         } else
 166                 dsos__add(kernel_dso);
 167
 168         vdso = dso__new("[vdso]", 0);
 169         if (!vdso)
 170                 return -1;
 171
 172         vdso->find_symbol = vdso__find_symbol;
 173
 174         dsos__add(vdso);
 175
 176         return err;
 177 }
 178
 179 static char __cwd[PATH_MAX];
 180 static char *cwd = __cwd;
 181 static int cwdlen;
 182
 183 static int strcommon(const char *pathname)
 184 {
 185         int n = 0;
 186
 187         while (pathname[n] == cwd[n] && n < cwdlen)
 188                 ++n;
 189
 190         return n;
 191 }
 192
 193 struct map {
 194         struct list_head node;
 195         uint64_t         start;
 196         uint64_t         end;
 197         uint64_t         pgoff;
 198         uint64_t         (*map_ip)(struct map *, uint64_t);
 199         struct dso       *dso;
 200 };
 201
 202 static uint64_t map__map_ip(struct map *map, uint64_t ip)
 203 {
 204         return ip - map->start + map->pgoff;
 205 }
 206
 207 static uint64_t vdso__map_ip(struct map *map, uint64_t ip)
 208 {
 209         return ip;
 210 }
 211
 212 static inline int is_anon_memory(const char *filename)
 213 {
 214      return strcmp(filename, "//anon") == 0;
 215 }
 216
 217 static struct map *map__new(struct mmap_event *event)
 218 {
 219         struct map *self = malloc(sizeof(*self));
 220
 221         if (self != NULL) {
 222                 const char *filename = event->filename;
 223                 char newfilename[PATH_MAX];
 224                 int anon;
 225
 226                 if (cwd) {
 227                         int n = strcommon(filename);
 228
 229                         if (n == cwdlen) {
 230                                 snprintf(newfilename, sizeof(newfilename),
 231                                          ".%s", filename + n);
 232                                 filename = newfilename;
 233                         }
 234                 }
 235
 236                 anon = is_anon_memory(filename);
 237
 238                 if (anon) {
 239                         snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
 240                         filename = newfilename;
 241                 }
 242
 243                 self->start = event->start;
 244                 self->end   = event->start + event->len;
 245                 self->pgoff = event->pgoff;
 246
 247                 self->dso = dsos__findnew(filename);
 248                 if (self->dso == NULL)
 249                         goto out_delete;
 250
 251                 if (self->dso == vdso || anon)
 252                         self->map_ip = vdso__map_ip;
 253                 else
 254                         self->map_ip = map__map_ip;
 255         }
 256         return self;
 257 out_delete:
 258         free(self);
 259         return NULL;
 260 }
 261
 262 static struct map *map__clone(struct map *self)
 263 {
 264         struct map *map = malloc(sizeof(*self));
 265
 266         if (!map)
 267                 return NULL;
 268
 269         memcpy(map, self, sizeof(*self));
 270
 271         return map;
 272 }
 273
 274 static int map__overlap(struct map *l, struct map *r)
 275 {
 276         if (l->start > r->start) {
 277                 struct map *t = l;
 278                 l = r;
 279                 r = t;
 280         }
 281
 282         if (l->end > r->start)
 283                 return 1;
 284
 285         return 0;
 286 }
 287
 288 static size_t map__fprintf(struct map *self, FILE *fp)
 289 {
 290         return fprintf(fp, " %"PRIx64"-%"PRIx64" %"PRIx64" %s\n",
 291                        self->start, self->end, self->pgoff, self->dso->name);
 292 }
 293
 294
 295 struct thread {
 296         struct rb_node   rb_node;
 297         struct list_head maps;
 298         pid_t            pid;
 299         char             *comm;
 300 };
 301
 302 static struct thread *thread__new(pid_t pid)
 303 {
 304         struct thread *self = malloc(sizeof(*self));
 305
 306         if (self != NULL) {
 307                 self->pid = pid;
 308                 self->comm = malloc(32);
 309                 if (self->comm)
 310                         snprintf(self->comm, 32, ":%d", self->pid);
 311                 INIT_LIST_HEAD(&self->maps);
 312         }
 313
 314         return self;
 315 }
 316
 317 static int thread__set_comm(struct thread *self, const char *comm)
 318 {
 319         if (self->comm)
 320                 free(self->comm);
 321         self->comm = strdup(comm);
 322         return self->comm ? 0 : -ENOMEM;
 323 }
 324
 325 static size_t thread__fprintf(struct thread *self, FILE *fp)
 326 {
 327         struct map *pos;
 328         size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
 329
 330         list_for_each_entry(pos, &self->maps, node)
 331                 ret += map__fprintf(pos, fp);
 332
 333         return ret;
 334 }
 335
 336
 337 static struct rb_root threads;
 338 static struct thread *last_match;
 339
 340 static struct thread *threads__findnew(pid_t pid)
 341 {
 342         struct rb_node **p = &threads.rb_node;
 343         struct rb_node *parent = NULL;
 344         struct thread *th;
 345
 346         /*
 347          * Font-end cache - PID lookups come in blocks,
 348          * so most of the time we dont have to look up
 349          * the full rbtree:
 350          */
 351         if (last_match && last_match->pid == pid)
 352                 return last_match;
 353
 354         while (*p != NULL) {
 355                 parent = *p;
 356                 th = rb_entry(parent, struct thread, rb_node);
 357
 358                 if (th->pid == pid) {
 359                         last_match = th;
 360                         return th;
 361                 }
 362
 363                 if (pid < th->pid)
 364                         p = &(*p)->rb_left;
 365                 else
 366                         p = &(*p)->rb_right;
 367         }
 368
 369         th = thread__new(pid);
 370         if (th != NULL) {
 371                 rb_link_node(&th->rb_node, parent, p);
 372                 rb_insert_color(&th->rb_node, &threads);
 373                 last_match = th;
 374         }
 375
 376         return th;
 377 }
 378
 379 static void thread__insert_map(struct thread *self, struct map *map)
 380 {
 381         struct map *pos, *tmp;
 382
 383         list_for_each_entry_safe(pos, tmp, &self->maps, node) {
 384                 if (map__overlap(pos, map)) {
 385                         list_del_init(&pos->node);
 386                         /* XXX leaks dsos */
 387                         free(pos);
 388                 }
 389         }
 390
 391         list_add_tail(&map->node, &self->maps);
 392 }
 393
 394 static int thread__fork(struct thread *self, struct thread *parent)
 395 {
 396         struct map *map;
 397
 398         if (self->comm)
 399                 free(self->comm);
 400         self->comm = strdup(parent->comm);
 401         if (!self->comm)
 402                 return -ENOMEM;
 403
 404         list_for_each_entry(map, &parent->maps, node) {
 405                 struct map *new = map__clone(map);
 406                 if (!new)
 407                         return -ENOMEM;
 408                 thread__insert_map(self, new);
 409         }
 410
 411         return 0;
 412 }
 413
 414 static struct map *thread__find_map(struct thread *self, uint64_t ip)
 415 {
 416         struct map *pos;
 417
 418         if (self == NULL)
 419                 return NULL;
 420
 421         list_for_each_entry(pos, &self->maps, node)
 422                 if (ip >= pos->start && ip <= pos->end)
 423                         return pos;
 424
 425         return NULL;
 426 }
 427
 428 static size_t threads__fprintf(FILE *fp)
 429 {
 430         size_t ret = 0;
 431         struct rb_node *nd;
 432
 433         for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
 434                 struct thread *pos = rb_entry(nd, struct thread, rb_node);
 435
 436                 ret += thread__fprintf(pos, fp);
 437         }
 438
 439         return ret;
 440 }
 441
 442 /*
 443  * histogram, sorted on item, collects counts
 444  */
 445
 446 static struct rb_root hist;
 447
 448 struct hist_entry {
 449         struct rb_node   rb_node;
 450
 451         struct thread    *thread;
 452         struct map       *map;
 453         struct dso       *dso;
 454         struct symbol    *sym;
 455         uint64_t         ip;
 456         char             level;
 457
 458         uint32_t         count;
 459 };
 460
 461 /*
 462  * configurable sorting bits
 463  */
 464
 465 struct sort_entry {
 466         struct list_head list;
 467
 468         char *header;
 469
 470         int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
 471         int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
 472         size_t  (*print)(FILE *fp, struct hist_entry *);
 473 };
 474
 475 /* --sort pid */
 476
 477 static int64_t
 478 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
 479 {
 480         return right->thread->pid - left->thread->pid;
 481 }
 482
 483 static size_t
 484 sort__thread_print(FILE *fp, struct hist_entry *self)
 485 {
 486         return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
 487 }
 488
 489 static struct sort_entry sort_thread = {
 490         .header = "         Command:  Pid",
 491         .cmp    = sort__thread_cmp,
 492         .print  = sort__thread_print,
 493 };
 494
 495 /* --sort comm */
 496
 497 static int64_t
 498 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
 499 {
 500         return right->thread->pid - left->thread->pid;
 501 }
 502
 503 static int64_t
 504 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
 505 {
 506         char *comm_l = left->thread->comm;
 507         char *comm_r = right->thread->comm;
 508
 509         if (!comm_l || !comm_r) {
 510                 if (!comm_l && !comm_r)
 511                         return 0;
 512                 else if (!comm_l)
 513                         return -1;
 514                 else
 515                         return 1;
 516         }
 517
 518         return strcmp(comm_l, comm_r);
 519 }
 520
 521 static size_t
 522 sort__comm_print(FILE *fp, struct hist_entry *self)
 523 {
 524         return fprintf(fp, "%16s", self->thread->comm);
 525 }
 526
 527 static struct sort_entry sort_comm = {
 528         .header         = "         Command",
 529         .cmp            = sort__comm_cmp,
 530         .collapse       = sort__comm_collapse,
 531         .print          = sort__comm_print,
 532 };
 533
 534 /* --sort dso */
 535
 536 static int64_t
 537 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
 538 {
 539         struct dso *dso_l = left->dso;
 540         struct dso *dso_r = right->dso;
 541
 542         if (!dso_l || !dso_r) {
 543                 if (!dso_l && !dso_r)
 544                         return 0;
 545                 else if (!dso_l)
 546                         return -1;
 547                 else
 548                         return 1;
 549         }
 550
 551         return strcmp(dso_l->name, dso_r->name);
 552 }
 553
 554 static size_t
 555 sort__dso_print(FILE *fp, struct hist_entry *self)
 556 {
 557         if (self->dso)
 558                 return fprintf(fp, "%-25s", self->dso->name);
 559
 560         return fprintf(fp, "%016llx         ", (__u64)self->ip);
 561 }
 562
 563 static struct sort_entry sort_dso = {
 564         .header = "Shared Object            ",
 565         .cmp    = sort__dso_cmp,
 566         .print  = sort__dso_print,
 567 };
 568
 569 /* --sort symbol */
 570
 571 static int64_t
 572 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
 573 {
 574         uint64_t ip_l, ip_r;
 575
 576         if (left->sym == right->sym)
 577                 return 0;
 578
 579         ip_l = left->sym ? left->sym->start : left->ip;
 580         ip_r = right->sym ? right->sym->start : right->ip;
 581
 582         return (int64_t)(ip_r - ip_l);
 583 }
 584
 585 static size_t
 586 sort__sym_print(FILE *fp, struct hist_entry *self)
 587 {
 588         size_t ret = 0;
 589
 590         if (verbose)
 591                 ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
 592
 593         if (self->sym) {
 594                 ret += fprintf(fp, "[%c] %s",
 595                         self->dso == kernel_dso ? 'k' : '.', self->sym->name);
 596         } else {
 597                 ret += fprintf(fp, "%#016llx", (__u64)self->ip);
 598         }
 599
 600         return ret;
 601 }
 602
 603 static struct sort_entry sort_sym = {
 604         .header = "Symbol",
 605         .cmp    = sort__sym_cmp,
 606         .print  = sort__sym_print,
 607 };
 608
 609 static int sort__need_collapse = 0;
 610
 611 struct sort_dimension {
 612         char                    *name;
 613         struct sort_entry       *entry;
 614         int                     taken;
 615 };
 616
 617 static struct sort_dimension sort_dimensions[] = {
 618         { .name = "pid",        .entry = &sort_thread,  },
 619         { .name = "comm",       .entry = &sort_comm,    },
 620         { .name = "dso",        .entry = &sort_dso,     },
 621         { .name = "symbol",     .entry = &sort_sym,     },
 622 };
 623
 624 static LIST_HEAD(hist_entry__sort_list);
 625
 626 static int sort_dimension__add(char *tok)
 627 {
 628         int i;
 629
 630         for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
 631                 struct sort_dimension *sd = &sort_dimensions[i];
 632
 633                 if (sd->taken)
 634                         continue;
 635
 636                 if (strncasecmp(tok, sd->name, strlen(tok)))
 637                         continue;
 638
 639                 if (sd->entry->collapse)
 640                         sort__need_collapse = 1;
 641
 642                 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
 643                 sd->taken = 1;
 644
 645                 return 0;
 646         }
 647
 648         return -ESRCH;
 649 }
 650
 651 static int64_t
 652 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
 653 {
 654         struct sort_entry *se;
 655         int64_t cmp = 0;
 656
 657         list_for_each_entry(se, &hist_entry__sort_list, list) {
 658                 cmp = se->cmp(left, right);
 659                 if (cmp)
 660                         break;
 661         }
 662
 663         return cmp;
 664 }
 665
 666 static int64_t
 667 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
 668 {
 669         struct sort_entry *se;
 670         int64_t cmp = 0;
 671
 672         list_for_each_entry(se, &hist_entry__sort_list, list) {
 673                 int64_t (*f)(struct hist_entry *, struct hist_entry *);
 674
 675                 f = se->collapse ?: se->cmp;
 676
 677                 cmp = f(left, right);
 678                 if (cmp)
 679                         break;
 680         }
 681
 682         return cmp;
 683 }
 684
 685 static size_t
 686 hist_entry__fprintf(FILE *fp, struct hist_entry *self, uint64_t total_samples)
 687 {
 688         struct sort_entry *se;
 689         size_t ret;
 690
 691         if (total_samples) {
 692                 double percent = self->count * 100.0 / total_samples;
 693                 char *color = PERF_COLOR_NORMAL;
 694
 695                 /*
 696                  * We color high-overhead entries in red, low-overhead
 697                  * entries in green - and keep the middle ground normal:
 698                  */
 699                 if (percent >= 5.0)
 700                         color = PERF_COLOR_RED;
 701                 if (percent < 0.5)
 702                         color = PERF_COLOR_GREEN;
 703
 704                 ret = color_fprintf(fp, color, "   %6.2f%%",
 705                                 (self->count * 100.0) / total_samples);
 706         } else
 707                 ret = fprintf(fp, "%12d ", self->count);
 708
 709         list_for_each_entry(se, &hist_entry__sort_list, list) {
 710                 fprintf(fp, "  ");
 711                 ret += se->print(fp, self);
 712         }
 713
 714         ret += fprintf(fp, "\n");
 715
 716         return ret;
 717 }
 718
 719 /*
 720  * collect histogram counts
 721  */
 722
 723 static int
 724 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
 725                 struct symbol *sym, uint64_t ip, char level)
 726 {
 727         struct rb_node **p = &hist.rb_node;
 728         struct rb_node *parent = NULL;
 729         struct hist_entry *he;
 730         struct hist_entry entry = {
 731                 .thread = thread,
 732                 .map    = map,
 733                 .dso    = dso,
 734                 .sym    = sym,
 735                 .ip     = ip,
 736                 .level  = level,
 737                 .count  = 1,
 738         };
 739         int cmp;
 740
 741         while (*p != NULL) {
 742                 parent = *p;
 743                 he = rb_entry(parent, struct hist_entry, rb_node);
 744
 745                 cmp = hist_entry__cmp(&entry, he);
 746
 747                 if (!cmp) {
 748                         he->count++;
 749                         return 0;
 750                 }
 751
 752                 if (cmp < 0)
 753                         p = &(*p)->rb_left;
 754                 else
 755                         p = &(*p)->rb_right;
 756         }
 757
 758         he = malloc(sizeof(*he));
 759         if (!he)
 760                 return -ENOMEM;
 761         *he = entry;
 762         rb_link_node(&he->rb_node, parent, p);
 763         rb_insert_color(&he->rb_node, &hist);
 764
 765         return 0;
 766 }
 767
 768 static void hist_entry__free(struct hist_entry *he)
 769 {
 770         free(he);
 771 }
 772
 773 /*
 774  * collapse the histogram
 775  */
 776
 777 static struct rb_root collapse_hists;
 778
 779 static void collapse__insert_entry(struct hist_entry *he)
 780 {
 781         struct rb_node **p = &collapse_hists.rb_node;
 782         struct rb_node *parent = NULL;
 783         struct hist_entry *iter;
 784         int64_t cmp;
 785
 786         while (*p != NULL) {
 787                 parent = *p;
 788                 iter = rb_entry(parent, struct hist_entry, rb_node);
 789
 790                 cmp = hist_entry__collapse(iter, he);
 791
 792                 if (!cmp) {
 793                         iter->count += he->count;
 794                         hist_entry__free(he);
 795                         return;
 796                 }
 797
 798                 if (cmp < 0)
 799                         p = &(*p)->rb_left;
 800                 else
 801                         p = &(*p)->rb_right;
 802         }
 803
 804         rb_link_node(&he->rb_node, parent, p);
 805         rb_insert_color(&he->rb_node, &collapse_hists);
 806 }
 807
 808 static void collapse__resort(void)
 809 {
 810         struct rb_node *next;
 811         struct hist_entry *n;
 812
 813         if (!sort__need_collapse)
 814                 return;
 815
 816         next = rb_first(&hist);
 817         while (next) {
 818                 n = rb_entry(next, struct hist_entry, rb_node);
 819                 next = rb_next(&n->rb_node);
 820
 821                 rb_erase(&n->rb_node, &hist);
 822                 collapse__insert_entry(n);
 823         }
 824 }
 825
 826 /*
 827  * reverse the map, sort on count.
 828  */
 829
 830 static struct rb_root output_hists;
 831
 832 static void output__insert_entry(struct hist_entry *he)
 833 {
 834         struct rb_node **p = &output_hists.rb_node;
 835         struct rb_node *parent = NULL;
 836         struct hist_entry *iter;
 837
 838         while (*p != NULL) {
 839                 parent = *p;
 840                 iter = rb_entry(parent, struct hist_entry, rb_node);
 841
 842                 if (he->count > iter->count)
 843                         p = &(*p)->rb_left;
 844                 else
 845                         p = &(*p)->rb_right;
 846         }
 847
 848         rb_link_node(&he->rb_node, parent, p);
 849         rb_insert_color(&he->rb_node, &output_hists);
 850 }
 851
 852 static void output__resort(void)
 853 {
 854         struct rb_node *next;
 855         struct hist_entry *n;
 856         struct rb_root *tree = &hist;
 857
 858         if (sort__need_collapse)
 859                 tree = &collapse_hists;
 860
 861         next = rb_first(tree);
 862
 863         while (next) {
 864                 n = rb_entry(next, struct hist_entry, rb_node);
 865                 next = rb_next(&n->rb_node);
 866
 867                 rb_erase(&n->rb_node, tree);
 868                 output__insert_entry(n);
 869         }
 870 }
 871
 872 static size_t output__fprintf(FILE *fp, uint64_t total_samples)
 873 {
 874         struct hist_entry *pos;
 875         struct sort_entry *se;
 876         struct rb_node *nd;
 877         size_t ret = 0;
 878
 879         fprintf(fp, "\n");
 880         fprintf(fp, "#\n");
 881         fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
 882         fprintf(fp, "#\n");
 883
 884         fprintf(fp, "# Overhead");
 885         list_for_each_entry(se, &hist_entry__sort_list, list)
 886                 fprintf(fp, "  %s", se->header);
 887         fprintf(fp, "\n");
 888
 889         fprintf(fp, "# ........");
 890         list_for_each_entry(se, &hist_entry__sort_list, list) {
 891                 int i;
 892
 893                 fprintf(fp, "  ");
 894                 for (i = 0; i < strlen(se->header); i++)
 895                         fprintf(fp, ".");
 896         }
 897         fprintf(fp, "\n");
 898
 899         fprintf(fp, "#\n");
 900
 901         for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
 902                 pos = rb_entry(nd, struct hist_entry, rb_node);
 903                 ret += hist_entry__fprintf(fp, pos, total_samples);
 904         }
 905
 906         if (!strcmp(sort_order, default_sort_order)) {
 907                 fprintf(fp, "#\n");
 908                 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
 909                 fprintf(fp, "#\n");
 910         }
 911         fprintf(fp, "\n");
 912
 913         return ret;
 914 }
 915
 916 static void register_idle_thread(void)
 917 {
 918         struct thread *thread = threads__findnew(0);
 919
 920         if (thread == NULL ||
 921                         thread__set_comm(thread, "[idle]")) {
 922                 fprintf(stderr, "problem inserting idle task.\n");
 923                 exit(-1);
 924         }
 925 }
 926
 927 static unsigned long total = 0,
 928                      total_mmap = 0,
 929                      total_comm = 0,
 930                      total_fork = 0,
 931                      total_unknown = 0;
 932
 933 static int
 934 process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
 935 {
 936         char level;
 937         int show = 0;
 938         struct dso *dso = NULL;
 939         struct thread *thread = threads__findnew(event->ip.pid);
 940         uint64_t ip = event->ip.ip;
 941         struct map *map = NULL;
 942
 943         dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p\n",
 944                 (void *)(offset + head),
 945                 (void *)(long)(event->header.size),
 946                 event->header.misc,
 947                 event->ip.pid,
 948                 (void *)(long)ip);
 949
 950         dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
 951
 952         if (thread == NULL) {
 953                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 954                         event->header.type);
 955                 return -1;
 956         }
 957
 958         if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
 959                 show = SHOW_KERNEL;
 960                 level = 'k';
 961
 962                 dso = kernel_dso;
 963
 964                 dprintf(" ...... dso: %s\n", dso->name);
 965
 966         } else if (event->header.misc & PERF_EVENT_MISC_USER) {
 967
 968                 show = SHOW_USER;
 969                 level = '.';
 970
 971                 map = thread__find_map(thread, ip);
 972                 if (map != NULL) {
 973                         ip = map->map_ip(map, ip);
 974                         dso = map->dso;
 975                 } else {
 976                         /*
 977                          * If this is outside of all known maps,
 978                          * and is a negative address, try to look it
 979                          * up in the kernel dso, as it might be a
 980                          * vsyscall (which executes in user-mode):
 981                          */
 982                         if ((long long)ip < 0)
 983                                 dso = kernel_dso;
 984                 }
 985                 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
 986
 987         } else {
 988                 show = SHOW_HV;
 989                 level = 'H';
 990                 dprintf(" ...... dso: [hypervisor]\n");
 991         }
 992
 993         if (show & show_mask) {
 994                 struct symbol *sym = NULL;
 995
 996                 if (dso)
 997                         sym = dso->find_symbol(dso, ip);
 998
 999                 if (hist_entry__add(thread, map, dso, sym, ip, level)) {
1000                         fprintf(stderr,
1001                 "problem incrementing symbol count, skipping event\n");
1002                         return -1;
1003                 }
1004         }
1005         total++;
1006
1007         return 0;
1008 }
1009
1010 static int
1011 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1012 {
1013         struct thread *thread = threads__findnew(event->mmap.pid);
1014         struct map *map = map__new(&event->mmap);
1015
1016         dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1017                 (void *)(offset + head),
1018                 (void *)(long)(event->header.size),
1019                 event->mmap.pid,
1020                 (void *)(long)event->mmap.start,
1021                 (void *)(long)event->mmap.len,
1022                 (void *)(long)event->mmap.pgoff,
1023                 event->mmap.filename);
1024
1025         if (thread == NULL || map == NULL) {
1026                 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1027                 return 0;
1028         }
1029
1030         thread__insert_map(thread, map);
1031         total_mmap++;
1032
1033         return 0;
1034 }
1035
1036 static int
1037 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1038 {
1039         struct thread *thread = threads__findnew(event->comm.pid);
1040
1041         dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1042                 (void *)(offset + head),
1043                 (void *)(long)(event->header.size),
1044                 event->comm.comm, event->comm.pid);
1045
1046         if (thread == NULL ||
1047             thread__set_comm(thread, event->comm.comm)) {
1048                 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1049                 return -1;
1050         }
1051         total_comm++;
1052
1053         return 0;
1054 }
1055
1056 static int
1057 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1058 {
1059         struct thread *thread = threads__findnew(event->fork.pid);
1060         struct thread *parent = threads__findnew(event->fork.ppid);
1061
1062         dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1063                 (void *)(offset + head),
1064                 (void *)(long)(event->header.size),
1065                 event->fork.pid, event->fork.ppid);
1066
1067         if (!thread || !parent || thread__fork(thread, parent)) {
1068                 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1069                 return -1;
1070         }
1071         total_fork++;
1072
1073         return 0;
1074 }
1075
1076 static int
1077 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1078 {
1079         dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1080                 (void *)(offset + head),
1081                 (void *)(long)(event->header.size),
1082                 event->period.time,
1083                 event->period.id,
1084                 event->period.sample_period);
1085
1086         return 0;
1087 }
1088
1089 static int
1090 process_event(event_t *event, unsigned long offset, unsigned long head)
1091 {
1092         if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
1093                 return process_overflow_event(event, offset, head);
1094
1095         switch (event->header.type) {
1096         case PERF_EVENT_MMAP:
1097                 return process_mmap_event(event, offset, head);
1098
1099         case PERF_EVENT_COMM:
1100                 return process_comm_event(event, offset, head);
1101
1102         case PERF_EVENT_FORK:
1103                 return process_fork_event(event, offset, head);
1104
1105         case PERF_EVENT_PERIOD:
1106                 return process_period_event(event, offset, head);
1107         /*
1108          * We dont process them right now but they are fine:
1109          */
1110
1111         case PERF_EVENT_THROTTLE:
1112         case PERF_EVENT_UNTHROTTLE:
1113                 return 0;
1114
1115         default:
1116                 return -1;
1117         }
1118
1119         return 0;
1120 }
1121
1122 static int __cmd_report(void)
1123 {
1124         int ret, rc = EXIT_FAILURE;
1125         unsigned long offset = 0;
1126         unsigned long head = 0;
1127         struct stat stat;
1128         event_t *event;
1129         uint32_t size;
1130         char *buf;
1131
1132         register_idle_thread();
1133
1134         input = open(input_name, O_RDONLY);
1135         if (input < 0) {
1136                 fprintf(stderr, " failed to open file: %s", input_name);
1137                 if (!strcmp(input_name, "perf.data"))
1138                         fprintf(stderr, "  (try 'perf record' first)");
1139                 fprintf(stderr, "\n");
1140                 exit(-1);
1141         }
1142
1143         ret = fstat(input, &stat);
1144         if (ret < 0) {
1145                 perror("failed to stat file");
1146                 exit(-1);
1147         }
1148
1149         if (!stat.st_size) {
1150                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1151                 exit(0);
1152         }
1153
1154         if (load_kernel() < 0) {
1155                 perror("failed to load kernel symbols");
1156                 return EXIT_FAILURE;
1157         }
1158
1159         if (!full_paths) {
1160                 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1161                         perror("failed to get the current directory");
1162                         return EXIT_FAILURE;
1163                 }
1164                 cwdlen = strlen(cwd);
1165         } else {
1166                 cwd = NULL;
1167                 cwdlen = 0;
1168         }
1169 remap:
1170         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1171                            MAP_SHARED, input, offset);
1172         if (buf == MAP_FAILED) {
1173                 perror("failed to mmap file");
1174                 exit(-1);
1175         }
1176
1177 more:
1178         event = (event_t *)(buf + head);
1179
1180         size = event->header.size;
1181         if (!size)
1182                 size = 8;
1183
1184         if (head + event->header.size >= page_size * mmap_window) {
1185                 unsigned long shift = page_size * (head / page_size);
1186                 int ret;
1187
1188                 ret = munmap(buf, page_size * mmap_window);
1189                 assert(ret == 0);
1190
1191                 offset += shift;
1192                 head -= shift;
1193                 goto remap;
1194         }
1195
1196         size = event->header.size;
1197
1198         dprintf("%p [%p]: event: %d\n",
1199                         (void *)(offset + head),
1200                         (void *)(long)event->header.size,
1201                         event->header.type);
1202
1203         if (!size || process_event(event, offset, head) < 0) {
1204
1205                 dprintf("%p [%p]: skipping unknown header type: %d\n",
1206                         (void *)(offset + head),
1207                         (void *)(long)(event->header.size),
1208                         event->header.type);
1209
1210                 total_unknown++;
1211
1212                 /*
1213                  * assume we lost track of the stream, check alignment, and
1214                  * increment a single u64 in the hope to catch on again 'soon'.
1215                  */
1216
1217                 if (unlikely(head & 7))
1218                         head &= ~7ULL;
1219
1220                 size = 8;
1221         }
1222
1223         head += size;
1224
1225         if (offset + head < stat.st_size)
1226                 goto more;
1227
1228         rc = EXIT_SUCCESS;
1229         close(input);
1230
1231         dprintf("      IP events: %10ld\n", total);
1232         dprintf("    mmap events: %10ld\n", total_mmap);
1233         dprintf("    comm events: %10ld\n", total_comm);
1234         dprintf("    fork events: %10ld\n", total_fork);
1235         dprintf(" unknown events: %10ld\n", total_unknown);
1236
1237         if (dump_trace)
1238                 return 0;
1239
1240         if (verbose >= 3)
1241                 threads__fprintf(stdout);
1242
1243         if (verbose >= 2)
1244                 dsos__fprintf(stdout);
1245
1246         collapse__resort();
1247         output__resort();
1248         output__fprintf(stdout, total);
1249
1250         return rc;
1251 }
1252
1253 static const char * const report_usage[] = {
1254         "perf report [<options>] <command>",
1255         NULL
1256 };
1257
1258 static const struct option options[] = {
1259         OPT_STRING('i', "input", &input_name, "file",
1260                     "input file name"),
1261         OPT_BOOLEAN('v', "verbose", &verbose,
1262                     "be more verbose (show symbol address, etc)"),
1263         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1264                     "dump raw trace in ASCII"),
1265         OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1266         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1267                    "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
1268         OPT_BOOLEAN('P', "full-paths", &full_paths,
1269                     "Don't shorten the pathnames taking into account the cwd"),
1270         OPT_END()
1271 };
1272
1273 static void setup_sorting(void)
1274 {
1275         char *tmp, *tok, *str = strdup(sort_order);
1276
1277         for (tok = strtok_r(str, ", ", &tmp);
1278                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1279                 if (sort_dimension__add(tok) < 0) {
1280                         error("Unknown --sort key: `%s'", tok);
1281                         usage_with_options(report_usage, options);
1282                 }
1283         }
1284
1285         free(str);
1286 }
1287
1288 int cmd_report(int argc, const char **argv, const char *prefix)
1289 {
1290         symbol__init();
1291
1292         page_size = getpagesize();
1293
1294         argc = parse_options(argc, argv, options, report_usage, 0);
1295
1296         setup_sorting();
1297
1298         /*
1299          * Any (unrecognized) arguments left?
1300          */
1301         if (argc)
1302                 usage_with_options(report_usage, options);
1303
1304         setup_pager();
1305
1306         return __cmd_report();
1307 }